dog

dogs

wug

wugs

toose

tooses

goose

geese

Background

How do children learn generalizations and exceptions?

  • Past-tense debate:
    • symbolic rules or connectionist networks?
    • dual-mechanism or single-mechanism?
  • Relationship between vocabulary and morphology as evidence
    • Marcus 1992: “overregularization does not correlate with […] children’s vocabularies”
    • Marchman 1992: continuity between vocabulary and morphology

Our approach:

  • Empirical landscape of vocabulary ↔️ morphology ↔️ age
  • Generalization across (at least a few) languages
  • Warning: 🚧



Parent-report checklist

  • “Action Words” → verbs
  • “Word endings” → correct inflections of irregulars
  • “Word forms” → overregularizations



Open repository of CDI data
wordbank.stanford.edu

  • Morphology data:
    • Danish
    • Norwegian
    • English (American)
    • English (Australian)

Overview

How does morphology…

  • change with age?
  • depend on vocabulary size?
  • depend on vocabulary size non-linearly?
  • depend on vocabulary size differently at different ages?
  • differ for different verbs?

Roadmap:

  • Data
  • Model selection
  • Results: overall effects
  • Results: effects for individual items
  • Takeaways

Data

  • Children who inflect at least one item
  • Items that have stem, correct form, overregularized form(s)
  • Code each child on each item:
    • stem only (says go, doesn’t say went/goed/wented)
    • stem + correct (says go and went, doesn’t say goed/wented)
    • stem + overregularized (says go and goed/wented)
  • Also each child’s number of verbs producing and age

Data

Data

Data

Data

Data

Data

Model selection

For each language and measure:

value ~ 1 + (1|stem) null

value ~ age + (1|stem) age only

value ~ verbs + (1|stem) verbs only

value ~ age + verbs + (1|stem) age and verbs

value ~ age * verbs + (1|stem) age, verbs, and their interaction

value ~ age + verbs + verbs^2 + (1|stem) age, verbs, and verbs quadratically

value ~ age * verbs + age * verbs^2 + (1|stem)
age, verbs, verbs quadratically, and age-verbs interactions

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

Model selection

main effects: age + verbs + verbs^2
interactions: age * verbs + age * verbs^2

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem only: says go, doesn’t say went/goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + correct: says go and went, doesn’t say goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: overall effects

stem + overregularized: says go and goed/wented

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

value ~ age + verbs + verbs^2 + (1 + age + verbs | stem)

Results: individual items

Results: individual items

verbs that are correctly inflected more are overregularized more

Results: individual items

Results: individual items

for verbs for which inflection is more influenced by vocabulary size, overregularization is also more influenced by vocabulary size

Results: individual items

Results: individual items

for verbs for which inflection is more influenced by age, overregularization is less influenced by age

Results: individual items

Results: individual items

verbs that are correctly inflected more are influenced by vocabulary size less

Results: individual items

Results: individual items

verbs that are overregularized more are influenced by vocabulary size less

Results: individual items

Results: individual items

verbs that are correctly inflected more are influenced by age less

Results: individual items

Results: individual items

verbs that are overregularized more are ???

Results: individual items

Results: individual items

for verbs for which inflection is influenced by vocabulary size more, inflection is influenced by age more

Results: individual items

Results: individual items

???

Takeaways

  • Irregular morphology learning relates strongly to vocabulary learning
  • This relationship is non-linear: as vocabulary increases…
    • vocabulary drives correct inflection more strongly
    • vocabulary drives overregularization less strongly
  • and is modulated by age: older children…
    • are more likely to produce correct forms and overregularize
    • are less strongly driven to correct inflection by vocabulary size
  • and varies across verbs: verbs that are correctly inflected more
    • are also overregularized more
    • are influenced more by vocabulary size and less by age

All data and code (🚧🚧🚧): github.com/mikabr/cdi-morph